AI Paper Atlas

Research map / Benchmarks / ImageNet

ImageNet leaderboard

1,688 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ImageNet. Best accuracy: 94.9 (SATA-B*, CVPR 2025).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ImageNet · accuracy

920 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
194.9SATA-B*
ViT-Base/16
SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers
CVPR 2025 · Table 1
Attention and transformers
294.2CALIP-FS
CLIP RN50
CALIP: Zero-Shot Enhancement of CLIP with Parameter-Free Attention
AAAI 2023 · Different Visual Encoders tabl
CLIP and zero-shot
392.7CE-CLIP+
CLIP ViT-B/32
Contrasting Intra-Modal and Ranking Cross-Modal Hard Negatives to Enhance Visio-Linguistic Compositional Understanding
CVPR 2024 · Table 5
CLIP and zero-shot
492.22SFT-M
Qwen2.5-VL-7B
Why Does RL Generalize Better Than SFT? A Data-Centric Perspective on VLM Post-Training
CVPR 2026 · Table 2
Visual and representation RL
591.95DART (MA)
ResNet-50
DART: Diversify-Aggregate-Repeat Training Improves Generalization of Neural Networks
CVPR 2023 · DART on ImageNet-1K and finegr
Domain adaptation and generalization
691.82Neural network regression
ResNet-50
Are Labels Always Necessary for Classifier Accuracy Evaluation?
CVPR 2021 · Table 1
Recognition and long-tail
791CSQ_D
ResNet-50
A Lower Bound of Hash Codes' Performance
NeurIPS 2022 · Recognition table
Prediction and decision losses
890.94Greedy soup
ViT-G/14
Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time
ICML 2022 · Table: ViT-G/14 results
Pruning and efficient tuning
990.7NCL + HNG
ResNet-18
Neighborhood Contrastive Learning for Novel Class Discovery
CVPR 2021 · Table 3
Recognition and long-tail
1090.45ViT-G/14Scaling Vision Transformers
CVPR 2022 · Table: ViT-G/14 results
Attention and transformers
910 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · accuracy (val)

501 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
191.8DeSparsify single-image
DeiT-s + A-ViT
DeSparsify: Adversarial Attack Against Token Sparsification Mechanisms
NeurIPS 2024 · Table 1
Attacks on text and graphs
289.7EVA
ViT giant
EVA: Exploring the Limits of Masked Visual Representation Learning at Scale
CVPR 2023 · Image classification compariso
Open-set and few-shot detection
389.2M3I Pre-training (InternImage-H)
InternImage-H (1B)
Towards All-in-One Pre-Training via Maximizing Multi-Modal Mutual Information
CVPR 2023 · Table: Pre-training of 1B Imag
CNN features
489.2NFNet-F4 (transfer)High-Performance Large-Scale Image Recognition Without Normalization
ICML 2021 · Appendix Table: Comparing Imag
RNNs and normalization
588.9ConvNeXt V2-H, FCMAE, IN-22K fine-tuned
ConvNeXt V2-H
ConvNeXt V2: Co-Designing and Scaling ConvNets With Masked Autoencoders
CVPR 2023 · ImageNet experiments, IN-22K r
CNN design
688.78DBMEF (ViTh-CLIP)
ViTh-CLIP
An Efficient Framework for Enhancing Discriminative Models via Diffusion Techniques
AAAI 2025 · Table 1
Fast diffusion sampling
788.5Gemini re-ranking
DinoV2 ViT-L14
Towards flexible perception with visual memory
ICML 2025 · text
Human and embodied modeling
888.4Finetuning with AdaVision tests
ViT-H/14
Adaptive Testing of Computer Vision Models
ICCV 2023 · Table 2
Human and embodied modeling
988.4Noisy Student Training (EfficientNet-L2)
EfficientNet-L2
Self-Training With Noisy Student Improves ImageNet Classification
CVPR 2020 · Table: Top-1 and Top-5 Accurac
Knowledge distillation
1088.2InternViT-6BInternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks
CVPR 2024 · Table 4
CLIP and zero-shot
491 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-5 accuracy

99 papers, each with its best reported number for its own method; higher is better.

#top-5 accuracy ↑MethodPaperTopic
199.8SATA-B*
ViT-Base/16
SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers
CVPR 2025 · Table 1
Attention and transformers
298.68Top-k DiffSortNets
Noisy Student EfficientNet-L2
Differentiable Top-k Classification Learning
ICML 2022 · Table 6
Recognition and long-tail
398.03S2-Boosting
ViT-L
S2-Boost: Synergistic Semantic Boosting for Coarse-to-Fine Ensemble Learning
AAAI 2026 · Table 1
Human and embodied modeling
497.1LSH-MoE
Swin-MoE
LSH-MoE: Communication-efficient MoE Training via Locality-Sensitive Hashing
NeurIPS 2024 · Swin-MoE results table
Mixture of experts
596.8Adaptive micro-batch clipping
DeiT-B
Revisit Micro-batch Clipping: Adaptive Data Pruning via Gradient Manipulation
ICLR 2025 · Table: ViT trained w/ or w/o a
Optimizers
696.43Swin-B-DIMAP1
Swin-B
Data-independent Module-aware Pruning for Hierarchical Vision Transformers
ICLR 2024 · Table 1
Attention and transformers
796.33ViT-B/16-224 + Vertical TM
ViT-B/16-224
TokenMixup: Efficient Attention-guided Token-level Data Augmentation for Transformers
NeurIPS 2022 · ImageNet results table
Attention and transformers
896PreNAS-BPreNAS: Preferred One-Shot Learning Towards Efficient Neural Architecture Search
ICML 2023 · Table: Comparison of different
Neural architecture search
995.87MambaVision-T + Token reordering
MambaVision-T
Demystifying the Token Dynamics of Deep Selective State Space Models
ICLR 2025 · Table: ImageNet-1K top-1 and t
Attention and transformers
1095.6SuperMix + RA
ResNet-200
SuperMix: Supervising the Mixing Data Augmentation
CVPR 2021 · Table 1
Knowledge distillation
89 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-1 error (val)

73 papers, each with its best reported number for its own method; lower is better.

#top-1 error ↓MethodPaperTopic
114.87Res2Net-200-SSLD
Res2Net-200
Res2Net: A New Multi-Scale Backbone Architecture
TPAMI 2019 · Table 1
CNN features
218.3PVT-LargePyramid Vision Transformer: A Versatile Backbone for Dense Prediction Without Convolutions
ICCV 2021 · Table 2
CNN features
318.68SENet + last-layer Dropout (drop ratio 0.2)
SENet
Understanding the Disharmony Between Dropout and Batch Normalization by Variance Shift
CVPR 2019 · ImageNet last-layer Dropout ta
RNNs and normalization
419.3OHL-Auto-Aug
SE-ResNeXt-101
Online Hyper-Parameter Learning for Auto-Augmentation Strategy
ICCV 2019 · Table 3
Neural architecture search
519.34AOGNet-60MAOGNets: Compositional Grammatical Architectures for Deep Learning
CVPR 2019 · ImageNet-1K results table
CNN features
619.41WR-41 + drop path
WR-41
Naive-Student: Leveraging Semi-Supervised Learning in Video Sequences for Urban Scene Segmentation
ECCV 2020 · Table 6
Semi-supervised learning
719.47CutMix
ResNeXt-101
CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features
ICCV 2019 · Table 3
Weakly supervised detection and segmentation
819.64Inception-v4-Gated
Inception-v4
You Look Twice: GaterNet for Dynamic Filter Selection in CNNs
CVPR 2019 · Table 3
CNN design
919.8mixup α=0.4
ResNeXt-101 64×4d
mixup: Beyond Empirical Risk Minimization
ICLR 2018 · Table 1
Certified and empirical robustness
1019.97VCR with dynamic attention sampler
ResNeXt-50 + VCR
Visual Concept Reasoning Networks
AAAI 2021 · Ablation Table (a)
Open-set and few-shot detection
63 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-5 accuracy (val)

68 papers, each with its best reported number for its own method; higher is better.

58 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-1 error

54 papers, each with its best reported number for its own method; lower is better.

#top-1 error ↓MethodPaperTopic
116.42DASH searched ConvNeXt-T
ConvNeXt-T
Efficient Architecture Search for Diverse Tasks
NeurIPS 2022 · Appendix table
Neural architecture search
218.4ACON-C
SENet-154
Activate or Not: Learning Customized Activation
CVPR 2021 · Table: Comparison on the extre
CNN features
318.87MCMC-Aug
ResNet-200
A Continuous Mapping For Augmentation Design
NeurIPS 2021 · Table 2
Optimizers
419.4CE-Net-G2CE-NAS: An End-to-End Carbon-Efficient Neural Architecture Search Framework
NeurIPS 2024 · ImageNet search results table
Neural architecture search
520.09ResNet-101 + SaliencyMix
ResNet-101
SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization
ICLR 2021 · Table 2
CNN features
620.45CAM-loss
ResNext-101 8x32d
Towards Learning Spatially Discriminative Feature Representations
ICCV 2021 · Table 1
Knowledge distillation
720.75Hard PatchUpPatchUp: A Feature-Space Block-Level Regularization Technique for Convolutional Neural Networks
AAAI 2022 · Table: Classification error ra
Certified and empirical robustness
820.77SENet+ABN
SENet
Attention Branch Network: Learning of Attention Mechanism for Visual Explanation
CVPR 2019 · Table: Comparison of top-1 err
Human and embodied modeling
920.93Snapshot Distillation, ResNet152
ResNet152
Snapshot Distillation: Teacher-Student Optimization in One Generation
CVPR 2019 · Table 2
CNN features
1021.1DKE-ResNet-101Differentiable Kernel Evolution
ICCV 2019 · Table 1
CNN features
44 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · FID

41 papers, each with its best reported number for its own method; lower is better.

#FID ↓MethodPaperTopic
12DLC₅₁₂ unconditional
DiT-XL/2
Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models
NeurIPS 2025 · Table: unconditional and condi
Autoregressive and latent generation
22.1COAP
SiT-XL/2
COAP: Memory-Efficient Training with Correlation-Aware Gradient Projection
CVPR 2025 · Table 2
Pruning and efficient tuning
32.55DDPM w/ proposed regularization
DDPM
On Error Propagation of Diffusion Models
ICLR 2024 · Table 1
Fast diffusion sampling
42.58ToCa (N=4, R=93%)
DiT-XL/2-G
Accelerating Diffusion Transformers with Token-wise Feature Caching
ICLR 2025 · Table 3
Fast diffusion sampling
52.63VAR-d24 + CCA
VAR-d24
Toward Guidance-Free AR Visual Generation via Condition Contrastive Alignment
ICLR 2025 · Table 2
Multimodal LLMs
62.86IG Ours
DiT-XL/2
Cross-fluctuation phase transitions reveal sampling dynamics in diffusion models
NeurIPS 2025 · Table, Class conditional gener
Fast diffusion sampling
73.1Nested models
MaGNeTS
Masked Generative Nested Transformers with Decode Time Scaling
ICML 2025 · Nested vs Standalone Models ta
Attention and transformers
83.9SSAH
ResNet-50
Frequency-Driven Imperceptible Adversarial Attack on Semantic Similarity
CVPR 2022 · Table 1
Attacks on text and graphs
95.32VQ-Diffusion (acc0.05)Vector Quantized Diffusion Model for Text-to-Image Synthesis
CVPR 2022 · Table: FID score comparison fo
Text-to-image diffusion
105.492DiffUEDiffUE: Enhancing Utility-Unlearnability Trade-off of Unlearnable Examples via Diffusion Autoencoders
ECCV 2026 · Table 4
Autoregressive and latent generation
31 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-5 error (%) (val)

38 papers, each with its best reported number for its own method; lower is better.

#top-5 error (%) ↓MethodPaperTopic
14.47SENet + last-layer Dropout (drop ratio 0.2)
SENet
Understanding the Disharmony Between Dropout and Batch Normalization by Variance Shift
CVPR 2019 · ImageNet last-layer Dropout ta
RNNs and normalization
24.57OHL-Auto-Aug
SE-ResNeXt-101
Online Hyper-Parameter Learning for Auto-Augmentation Strategy
ICCV 2019 · Table 3
Neural architecture search
34.78AOGNet-60MAOGNets: Compositional Grammatical Architectures for Deep Learning
CVPR 2019 · ImageNet-1K results table
CNN features
44.8Inception-v4-Gated
Inception-v4
You Look Twice: GaterNet for Dynamic Filter Selection in CNNs
CVPR 2019 · Table 3
CNN design
54.9mixup α=0.4
ResNeXt-101 64×4d
mixup: Beyond Empirical Risk Minimization
ICLR 2018 · Table 1
Certified and empirical robustness
65.173G-Net
ResNet-101
Deep Global Generalized Gaussian Networks
CVPR 2019 · Table 3
CNN design
75.17Modified XceptionEncoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
ECCV 2018 · Table 4
CNN features
85.2Evolutionary search, hierarchical representation, c0=64Hierarchical Representations for Efficient Architecture Search
ICLR 2018 · Table: Classification results
Neural architecture search
95.3ResNeXt-d101-c32 with SDPoint
ResNeXt-d101-c32
Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks
CVPR 2018 · Table 2
CNN design
105.57Our Inception-ResNet WITH SVB+BN
Inception-ResNet
Improving Training of Deep Neural Networks via Singular Value Bounding
CVPR 2017 · Table: ImageNet results
RNNs and normalization
28 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-5 error (%)

16 papers, each with its best reported number for its own method; lower is better.

#top-5 error (%) ↓MethodPaperTopic
14.03Relay BP
Inception-v3
Relay Backpropagation for Effective Learning of Deep Convolutional Neural Networks
ECCV 2016 · Table 6
CNN design
25.15ResNet-101 + SaliencyMix
ResNet-101
SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization
ICLR 2021 · Table 2
CNN features
35.55Snapshot Distillation, ResNet101
ResNet101
Snapshot Distillation: Teacher-Student Optimization in One Generation
CVPR 2019 · Table 2
CNN features
45.74ResNeXt-101 12×10d
ResNet-101 1×64d
Collegial Ensembles
NeurIPS 2020 · Table: Primal and dual results
RNNs and normalization
56.19Saliency Grafting
ResNet-50
Saliency Grafting: Innocuous Attribution-Guided Mixup with Calibrated Label Mixing
AAAI 2022 · Table: ImageNet
Recognition and long-tail
66.358Prodpoly-ResNet50
ResNet50
P-nets: Deep Polynomial Neural Networks
CVPR 2020 · Table 3
RNNs and normalization
76.46Block-QNN-S, N=3Practical Block-Wise Neural Network Architecture Generation
CVPR 2018 · Table 3
Neural architecture search
86.7LFM-PDARTS-CIFAR100
PDARTS (CIFAR100)
Learning from Mistakes – a Framework for Neural Architecture Search
AAAI 2022 · Table 3
Neural architecture search
97.5PGNAS-D-121
DenseNet-121
Posterior-Guided Neural Architecture Search
AAAI 2020 · Table 6
Neural architecture search
107.59Recasting (Rbs+D)
DenseNet-121
Network Recasting: A Universal Method for Network Architecture Transformation
AAAI 2019 · ILSVRC2012 recasting results t
Knowledge distillation
6 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · Acc@1

15 papers, each with its best reported number for its own method; higher is better.

#Acc@1 ↑MethodPaperTopic
188.64ALIGN
EfficientNet-L2
Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision
ICML 2021 · Table 4
CLIP and zero-shot
283.38Ties-CA
Swin-Transformer small
CAMEx: Curvature-aware Merging of Experts
ICLR 2025 · ImageNet-1k results table
Pruning and efficient tuning
383.2SMM-Base (SNN)
Base
SMM Transformer: Leveraging Spiking Neural Networks for Multimodal Tasks
ICML 2026 · Table 3
Spiking networks
483.2Differentiable hierarchical tokenization
ViT-B16
Differentiable Hierarchical Visual Tokenization
NeurIPS 2025 · Table 1
Human and embodied modeling
582.51Vmanba-T + DenoiseRep
Vmanba-T
DenoiseRep: Denoising Model for Representation Learning
NeurIPS 2024 · Table: Classification tasks
CNN features
681.4DeiT-S + SSTA (ours)
DeiT-S
Self-supervised Models are Good Teaching Assistants for Vision Transformers
ICML 2022 · Table 2
Knowledge distillation
780.9ViT-B + Aux
ViT-B
Frozen Language Models Are Gradient Coherence Rectifiers in Vision Transformers
AAAI 2025 · Table 4
Language, safety and interpretability
879.4I-stage + S-stage
DeiT-S
Zero-TPrune: Zero-Shot Token Pruning through Leveraging of the Attention Graph in Pre-Trained Transformers
CVPR 2024 · Table 1
Pruning and efficient tuning
979.38ScMoE (Pos-2)
SwinV2-MoE-S
Shortcut-connected Expert Parallelism for Accelerating Mixture of Experts
ICML 2025 · Table: ImageNet-1K accuracy an
Mixture of experts
1078Swin-T-ADN (TTTT)
Swin-T
Adaptive Depth Networks with Skippable Sub-Paths
NeurIPS 2024 · Table, efficient inference met
RNNs and normalization
5 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · error rate

15 papers, each with its best reported number for its own method; lower is better.

#error rate ↓MethodPaperTopic
115.47DALBG-T2T-ViT-14 (ours)Fair and Accurate Decision Making through Group-Aware Learning
ICML 2023 · Table 3
Fairness
217.55CBNN
EfficientNet-B3
Boost Neural Networks by Checkpoints
NeurIPS 2021 · Table 1
RNNs and normalization
319.7DDAS
ResNet-200
Direct Differentiable Augmentation Search
ICCV 2021 · Table 3
Neural architecture search
421.1ResNet-50 + LocalDrop
ResNet-50
LocalDrop: A Hybrid Regularization for Deep Neural Networks
TPAMI 2021 · Table 2
Kernels and regression theory
521.4Fixup-init + Mixup
ResNet-101
Fixup Initialization: Residual Learning Without Normalization
ICLR 2019 · ImageNet test results table
RNNs and normalization
621.58Zero-Warmup
Swin-Tiny
When Will Gradient Regularization Be Harmful?
ICML 2024 · TinyImageNet and ImageNet resu
Optimizers
721.75Ours(Det)
Wide-ResNet50
WLD-Reg: A Data-Dependent Within-Layer Diversity Regularizer
AAAI 2023 · Table 2
RNNs and normalization
824.32FRAGrow
ResNet
When to Grow? A Fitting Risk-Aware Policy for Layer Growing in Deep Neural Networks
AAAI 2024 · Main Results table
RNNs and normalization
924.4AUM
ResNet-50
Identifying Mislabeled Data using the Area Under the Margin Ranking
NeurIPS 2020 · Table 3
Noisy labels
1025.1AENIB
ViT-S/16
Enhancing Multiple Reliability Measures via Nuisance-Extended Information Bottleneck
CVPR 2023 · Table 3
Attacks on text and graphs
5 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ImageNet · top-1 test error (%)

13 papers, each with its best reported number for its own method; lower is better.

#top-1 test error (%) ↓MethodPaperTopic
122.8PAMKD-pcdarts-ImageNet
Pcdarts
Performance-Aware Mutual Knowledge Distillation for Improving Neural Architecture Search
CVPR 2022 · Table 3
Neural architecture search
223.26Mixer-B-W
MLP-Mixer
Understanding MLP-Mixer as a wide and sparse MLP
ICML 2024 · Table 1 (lower)
RNNs and normalization
323.3GEMNet-BNASGEM: Neural Architecture Search via Graph Embedding Method
AAAI 2021 · ImageNet results table
Graph algorithms and clustering
423.8VIM-NAS (searched on ImageNet)Learning Latent Architectural Distribution in Differentiable Neural Architecture Search via Variational Information Maximization
ICCV 2021 · Table 2
Neural architecture search
524SWAP-NAS
DARTS search space
SWAP-NAS: Sample-Wise Activation Patterns for Ultra-fast NAS
ICLR 2024 · ImageNet results table
Neural architecture search
624XNASXNAS: Neural Architecture Search with Expert Advice
NeurIPS 2019 · ImageNet results table
Neural architecture search
724.2TE-NAS + DAG
TE-NAS
Deep Architecture Connectivity Matters for Its Convergence: A Fine-Grained Analysis
NeurIPS 2022 · Table 1
Implicit bias and generalization
824.5H^s-DASUnchain the Search Space with Hierarchical Differentiable Architecture Search
AAAI 2021 · Table 2
Neural architecture search
924.9EG-NASEG-NAS: Neural Architecture Search with Fast Evolutionary Exploration
AAAI 2024 · Table 2
Optimizers
1025CATE-DNGO-LS (large budget)CATE: Computation-aware Neural Architecture Encoding with Transformers
ICML 2021 · Table 3
Neural architecture search
3 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

8 less reported leaderboards for ImageNet (other metrics or splits) are in the CSV and the API.

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.